This invention relates to a system for knowledge representation in a computer system together with Natural Language Interface, together with the ability to recognize, store and use patterns in the knowledge representation, together with computerized means for automatically transforming information in the knowledge representation into a multitude of documents or other human interpretable displays in a plurality of different formats or views, together with means for user interaction with the knowledge representation through the documents or displays.
The Knowledge Representation system defines a novel database engine constituting a method for modeling knowledge as a network of concepts and the relationships of each concept to other concepts. Each concept is represented as a record in the database which is identified by a unique record reference number. Relationships are represented by storing the unique record reference numbers of relevant records within the records comprising the knowledge representation database.
The records are organized into strata or levels of abstraction by means of storing relationships to other records within a given stratum. These relationships organize each level into hierarchies and a plurality of other relational networks which represent generally accepted ways of thinking about various fields of knowledge.
The concepts in lower levels of abstraction are related to those in higher levels by storing the unique record reference numbers of records of simple (more abstract) concepts within the structure of records representing more complex (more specific) concepts. Thus, more complex concepts are defined by the records whose reference numbers are stored within their record structure.
This invention further relates to a novel system for a Natural Language Interface (NLI) including a new method of parsing and evaluating input expressions in conjunction with the information in the Knowledge Representation Database.
The system includes a method for dynamically updating the Knowledge Representation database to "learn" through interaction with system users. This is accomplished by programmed means which automatically recognize, store, and utilize patterns in the knowledge representation database.
The system further includes a novel method for automatically transforming knowledge embodied in the Database into a plurality of human-perceivable documents expressing the knowledge.
User interaction with the Knowledge Representation through the Documents enables user interactions typical of Graphical User Interfaces (GUI), Hypertext, Object Linking Embedding (OLE), and context sensitive operation in a single unified user interface.
The present invention can be evaluated by referring to five major constituents which are illustrated in FIG. 1.
A) The data-base engine referred to herein as a "knowledge representation" system is most closely related to hierarchical and network databases. It differs from previously disclosed databases in that this invention employs a record system which embeds the database record numbers of a plurality of other records within a list stored in a particular record in a network of interrelated records. Each record is allowed to maintain an unbounded list of relationships to other records within the Knowledge Representation database.
System concepts are defined which allow the creation of levels of abstraction within the knowledge representation.
Fundamental relationships are the minimum set of relationships which each record is required to have in order to assure that the concept is related to the other concepts of the knowledge representation database. Each record is required to store the reference number of at least one record in a higher level of abstraction than that record, and every record on each level of abstraction is required to store at least one record reference number of another record on the same level in a relationship which is characterized by an attribute typical of the level.
The Knowledge Representation system further requires a separate, non-permanent database in which to assemble descriptions of particular records from the network of records in the primary Knowledge Representation database.
The retrieval system of an individual record and the records whose reference numbers are embedded in its substructure provides means for relationship inheritance and is without precedent.
B) The means for pattern recognition, storage and utilization employed in the invention are unique and provide novel means for machine "learning" by recognizing patterns in the relationships comprising the records, storing the patterns as relationships and utilizing the patterns in creating additional concepts and relationships in the Knowledge Representation Database.
C) The Natural Language Interface (NLI) is a novel system for enabling human interaction with the Knowledge Representation Database by using human language. The NLI system employs novel means for parsing input expressions in that it employs a sequential combination of a transitional network parser (related to an Augmented Transition Network) and a pattern matching parser (related to Parsing by differences lists methods). Both parsers are further unique in that they are based on the level of abstraction and fundamental relationships of a concept in the knowledge representation database instead of the human language grammatical function of the concept.
D) The method for automatically deriving a multitude of documents is based on abstracting standard documents or displays of knowledge into Views, Classes, and Types, wherein the Types define the vocabulary, the Class defines the grammar, and the View defines the connectivity and interaction, and is without precedent.
E) The user interface integrates the principal characteristics of GUI, Hypertext, OLE, and context sensitive programming in a single integrated user environment. This integration is without precedent, and is implemented in the invention by novel means which are significant improvements over the means currently known to the art for these technologies.
In this invention, the hypertext-like behavior is a consequence of the relationships stored at each record so that the view document has the feel of having hypertext implemented on a massive scale; every icon is a portal to all other view documents containing the concept and is also a portal to all related concepts. This contrasts with the Hypertext idiom which typically employs user defined locational links between documents which operate as an adjunct to static document representation.
The Knowledge Representation system further utilizes external relationships which enable the linking of records in the database to file or programs external to the Knowledge Representation Database.